arXiv cs.CL· Ziquan Wang, Yuxuan Nie, Jian Ma, Yingjie Li·· 13 小时前AI 评分24
研究：LLM 在基于检索的表格分类中何时优于距离加权投票
Beyond the Vote: When Do LLMs Add Value to Retrieval-Based Tabular Classification?
AI 导读
研究对比 4 个通用 LLM 与 Distance-KNN 在 19 个表格分类数据集上的表现，固定演示数量时扩大检索候选池可提升所有预测器准确率，但 Distance-KNN 在每个池规模下平均准确率最高。LLM 的优势集中在检索邻居标签共识度低的场景，且随模型而异；在特征空间距离最远的样本组中，所有 LLM 平均表现均落后于 Distance-KNN。
正文 · 原文
Abstract:Retrieval-based in-context learning (ICL) enables large language models (LLMs) to predict tabular labels from similar examples, but it remains unclear whether LLM inference improves on direct voting over those same examples. We address this question by comparing four general-purpose LLMs with distance-weighted nearest-neighbor voting (Distance-KNN) using identical retrieved neighbors across 19 tabular classification datasets. With the number of demonstrations fixed, enlarging the retrieval candidate pool improves accuracy for all predictors, yet Distance-KNN achieves the highest mean accuracy at every pool size. More detailed analyses reveal that LLM gains are concentrated where retrieved neighbors exhibit low label consensus and vary across models. These gains further depend on the average feature-space distance to the test query: under low label consensus, some LLMs outperform Distance-KNN in close and intermediate neighborhoods, but all fall behind on average in the most distant group. Complementary experiments on the number of demonstrations, prompt formulation, retrieval representation, and label-balanced demonstration selection largely preserve the overall findings. These results characterize when LLM inference adds predictive value beyond distance-weighted voting in retrieval-based tabular ICL.
| Subjects: | Information Retrieval (cs.IR); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.29520 [cs.IR] |
| (or arXiv:2610.29520v1 [cs.IR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.29520 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yingjie Li [view email]
[v1]
Wed, 21 Oct 2026 10:06:29 UTC (470 KB)
来源：arXiv cs.CL · arxiv.org